Phylogeography and genetic structure of the orchid Himantoglossum hircinum (L.) Spreng. across its European central–marginal gradient
Marion Pfeifer1*, Bertrand Schatz2, F. Xavier Pico´3, Nicodemo G. Passalacqua4, Michael F. Fay5, Pete D. Carey6and Florian Jeltsch1
1Institute of Vegetation Ecology and Nature Conservation, University of Potsdam, Maulbeerallee 2, Potsdam, Germany, 2Centre d’Ecologie Fonctionnelle et Evolutive, UMR
5175 Montpellier, France, 3Estacio´n Biolo´gica
de Don˜ana, CSIC, Avda. Mar´ıa Luisa s/n, Sevilla, Spain, 4Museo di Storia Naturale della Calabria ed Orto Botanico, Universita`della Calabria, 87030 Arcavacata di Rende, Cosenza, Italy, 5Jodrell Laboratory, Royal Botanic Gardens, Kew, Richmond, Surrey,
*Correspondence: Marion Pfeifer, Institute of Vegetation Ecology and Nature Conservation, University of Potsdam, Maulbeerallee 2, 14469
Aim This study aims to link demographic traits and post-glacial recolonization processes with genetic traits in Himantoglossum hircinum (L.) Spreng (Orchidaceae), and to test the implications of the central–marginal concept (CMC) in Europe.
Location Twenty sites covering the entire European distribution range of this species.
Methods We employed amplified fragment length polymorphism (AFLP) markers and performed a plastid microsatellite survey to assess genetic variation in 20 populations of H. hircinum located along central–marginal gradients. We measured demographic traits to assess population fitness along geographical gradients and to test for correlations between demographic traits and genetic diversity. We used genetic diversity indices and analyses of molecular variance (AMOVA) to test hypotheses of reduced genetic diversity and increased genetic differentiation and isolation from central to peripheral sites. We used Bayesian simulations to analyse genetic relationships among populations.
Results Genetic diversity decreased significantly with increasing latitudinal and longitudinal distance from the distribution centre when excluding outlying populations. The AMOVA revealed significant genetic differentiation among populations (FST = 0.146) and an increase in genetic differentiation from the centre of the geographical range to the margins (except for the Atlantic group). Population fitness, expressed as the ratio NR/N, decreased significantly with increasing latitudinal distance from the distribution centre. Flower production was lower in most eastern peripheral sites. The geographical distribution of microsatellite haplotypes suggests post-glacial range expansion along three major migratory pathways, as also supported by individual membership fractions in six ancestral genetic clusters (C1–C6). No correlations between genetic diversity (e.g. diversity indices, haplotype frequency) and population demographic traits were detected.
Main conclusions Reduced genetic diversity and haplotype frequency in H. hircinum at marginal sites reflect historical range expansions. Spatial variation in demographic traits could not explain genetic diversity patterns. For those sites that did not fit into the CMC, the genetic pattern is probably masked by other factors directly affecting either demography or population genetic structure. These include post-glacial recolonization patterns and changes in habitat suitability due to climate change at the northern periphery. Our findings emphasize the importance of distinguishing
historical effects from those caused by geographical variation in population demography of species when studying evolutionary and ecological processes at the range margins under global change.
Previous studies have tested the demographic assumptions of the ‘abundant centre’ model (or ‘central–marginal concept’, CMC) and its implications for geographical genetic structuring of species within Europe (reviewed by Sagarin & Gaines, 2002; Eckert et al., 2008). The model suggests that demographic and evolutionary processes at the margins of the range of a species determine not only its present-day distribution but also its response to changing environmental conditions (Kirkpatrick & Barton, 1997; Case & Taper, 2000). However, empirical evidence for model assumptions remains inconclusive, dem- onstrating that the CMC is far from fully understood (Eckert et al., 2008).
Declining environmental favourability towards the periph- ery of the geographical range (Hengeveld & Haeck, 1982) is predicted to result in a reduction in population size and density (Brown, 1984) and a decrease in both plant growth and reproduction (Parsons, 1991) towards range boundaries of species. However, whereas higher survival, fecundity and/or growth have been observed in central than in peripheral populations in some species, the opposite has been found in others (Sagarin & Gaines, 2002). Decline in population densities towards the high latitudinal distribution limits has been linked to reduced fecundity in peripheral populations of Decodon verticillatus (Dorken & Eckert, 2001), Cirsium acaule and Cirsium heterophyllum (Jump & Woodward, 2003) and greater interannual variation in life-state transition rates of passerine birds (Mehlman, 1997). However, reductions in fitness components do not necessarily have an impact on population growth and persistence, and thus fail to explain the position of range boundaries (Nantel & Gagnon, 1999; Kluth & Bruelheide, 2005). Also, geographical ranges examined in many species did not contract inwards when a species became endangered, and remnant populations of many species occurred exclusively in the periphery of the historical range (Channell & Lomolino, 2000).
The genetic implications of the CMC are numerous. Genetic diversity is hypothesized to decrease towards the range edge, because peripheral populations are expected to have lower effective population sizes and may consequently suffer from increased genetic drift and inbreeding (Giles & Goudet, 1997). Declining availability of suitable habitat patches and/or a higher probability of stochastic extinctions due to stronger demographic fluctuations towards range margins (Maurer &
Taper, 2002) are expected to increase isolation between populations. This, in combination with smaller population sizes, may lead to reduced gene flow and increased genetic differentiation (FST) between populations (Ellstrand & Hoff- man, 1990). This is relevant for understanding the distribution patterns of species, because a stochastic reduction in the genetic diversity of peripheral populations may limit their evolutionary potential, inhibiting the capacity of the species to adapt to conditions beyond the range limits (Blows & Hoffmann, 2005).
However, empirical studies show that patterns of population genetic structure across large spatial scales are variable and species-specific (Eckert et al., 2008). Peripheral populations may not show decreased genetic diversity (e.g. Cirsium heterophyllum, Jump & Woodward, 2003) or may contain genotypes evolved under variable, extreme and suboptimal conditions. Eckert et al. (2008) criticized many of the studies they reviewed for not considering historical influences in a phylogeographical framework and for not assessing any of the possible causes of reduced peripheral genetic diversity or greater differentiation, including estimates of population size, degree of spatial isolation among populations and demo- graphic turnover.
This study attempts to show how the distribution of species is shaped by historical processes, demographic traits and genetic processes interacting over geographical scales, and consequently how these processes determine the capacity of the species to respond to environmental changes. We studied fitness and the geographical genetic structure of populations of the lizard orchid, Himantoglossum hircinum (L.) Spreng. (Orchidaceae), along central–marginal gradients covering its entire distribution area in Europe, using amplified fragment length polymorphism (AFLP) markers and demographic monitoring. Plastid microsatellite markers were used to infer migratory pathways of the species and to assess the potential influence of historical processes on geographical genetic structure. Himantoglossum hircinum is a climate-sensitive plant that has recently experienced increases in abundance and population numbers at its north-eastern range margin, indicating a range shift (cf. Good, 1936; Carey, 1999). However, it is unclear whether these changes will compensate for the decline in abundance observed along the southern range margin of the species.
The following three questions were addressed in this study. (1) Can geographical trends be detected in population genetic
structure? (2) If geographical genetic structure is present, can the observed pattern be related to demographic traits (e.g. population fitness). (3) How important are historical processes for the current genetic structure of the species?
MATERIALS A ND M E THODS Study organism
Himantoglossum hircinum is a long-lived terrestrial orchid that perennates via tubers. Leaves emerge from below-ground tubers in late autumn and grow over the winter months; plants start to develop inflorescences in late April (Carey & Farrell,
2002). Population growth rate, transition probabilities between life stages and flowering probability of H. hircinum popula- tions at the north-eastern margin of its range have been shown to be strongly governed by variability in weather (Pfeifer et al.,
2006a,b). Flowering takes place between May and June, although a shift towards an earlier onset of flowering in some populations has been observed (M. Pfeifer, unpublished data). Plants are pollinated by generalist insects, precluding the possibility of range limitation due to an absence of pollinators (Bourne´rias & Prat, 2005). Seeds often remain lodged in the capsules, which are dropped beside the parent plant after maturation.
Study sites, measurements and sampling
Twenty sites distributed throughout the entire range of the species in Europe were sampled in 2007 (Table 1). The study covered three central sites, two sites in the northern range periphery, a number of sites along a gradient towards the eastern range margin and three sites in the southern range periphery of its sub-Mediterranean distribution (Fig. 1). Populations located in the mountains in northern Spain were grouped as Atlantic sites because of the likely influence of the Atlantic on their environment (e.g. weather conditions).
Population size (N) was estimated at each sample site in March. Leaf number was used as a measure of plant size, assuming that leaf number and summed leaf area are correlated (shown for 50 plants in 2004 at TH1: R2 = 0.87, P < 0.001). In June each year, populations were revisited and the number of reproductive plants (NR) was estimated. Because flowering and large plants contributed most to the finite rate of increase of the population (Pfeifer et al., 2006a), we used the ratio of NR/N and the number of flowers produced per plant as indicators of the condition of the population at each site. At each site, up to 16 flowering individuals were selected randomly for measurements and collection of three flowers each (Table 1). Plant size, inflorescence height and the number of flowers produced per plant were recorded. Fewer
Table 1 Traits of 20 populations of Himantoglossum hircinum sampled at five geographical regions within Europe. All measurements were carried out in May–June
2007. Geographical coordinates are given in decimal degrees (World Geodetic System 84). Bold type represents the presence of
long-term demographic monitoring plots.
Site Name Longitude Latitude Alt (m) N NR Ns Distribution centre
Alt, height above sea level (m); N and NR, estimates of population size and number of flowering plants (class 1, < 20 plants; class 2, 20–50 plants; class 3, 50–100 plants; class 4, 100–200 plants; class 5, 200–500 plants; class 6, 500–1000 plants; class 7, > 1000 plants); Ns, number of plants sampled per site.
Figure 1 Current distribution (light grey pattern, simplified) and major migration pathways (black arrows) of Himantoglossum hircinum in Europe during post-glacial expansion and more recent climate-induced northward range shifts. The dashed arrows indicate alternative pathways. The species probably survived in refugial zones (dark grey, encircled; see Schmitt, 2007) during the last ice age. R1 is the refugial zone of H. hircinum in southern France. R2 (Atlantic–Mediterranean element) and R3 (Adriatic–Mediterranean element) represent two major refugial and differentiation centres for many biota in southern Europe. The graphs show the percentage of haplotypes (H1 to H5) at the MSpsa1 locus in the sampled populations (black dots). Definitions of the abbreviations for population names are given in Table 1.
plants were sampled if the number of flowering plants was
< 10. In total, 211 plants were sampled. Flowers were stored in sealable plastic bags filled with silica gel (white, fine mesh size:
28–200 lm; Sigma-Aldrich Corp., Steinheim, Germany) to ensure rapid drying.
DNA extraction and AFLP analysis
DNA was extracted from dried flowers following a scaled- down 2 · cetyl trimethyl ammonium bromide (CTAB) extraction protocol based on Doyle & Doyle (1987) with minor modifications. DNA samples were cleaned using QIAquick columns (Qiagen, Valencia, CA, USA) and their quality was checked by running the extracts on a 1.0% agarose gel. The DNA concentration was determined using spectro- photometry. Following an initial screening of genetic variation in 16 plants from six different locations, two (EcoRI-ACG/
MseI-CAC, EcoRI-ACG/MseI-CTA) out of 17 AFLP primer combinations tested (see Appendix S1 in Supporting Infor- mation) appeared sufficiently polymorphic to discriminate clones within populations and were used in subsequent analyses of 205 plants (Table 2). The AFLP reactions (Vos et al., 1995) were conducted according to a modified AFLP® Plant Mapping Protocol of PE Applied Biosystems Inc. using EcoRI and MseI with 500 ng of isolated genomic DNA per sample. The success of each step was tested by running the PCR products on a 1.5% agarose gel (20 min, ±100 V). Fragments were run on an ABI Prism® 3100 Genetic Analyzer (Applied Biosystems Inc., Foster City, CA, USA) with 10 lL High Dye (deionized formamide) and 0.2 lL GeneScanTM500
ROXTM Size Standard per sample. Separated fragments were
Table 2 Genetic diversity and flowering performance computed for all sample sites of Himantoglossum hircinum in Europe (described in Table 1). A total of 205 plants were sampled. Genetic diversity (He) and differentiation (FST) differed among the five geographical regions. Linear modelling using the R 2.7.0 package showed that size had a significant positive effect on number of flowers produced per plant at most sites, while the strength of the relationship differed among sites (R2adj).
Site Ns %p PB He ± SE I ± SE S Mean(FPP) R2 Distribution centre: FST= 0.111, mean(He) = 0.176 ± 0.02
Eastern range margin: FST= 0.128, mean(He) = 0.137 ± 0.03
0.165 ± 0.016
0.105 ± 0.011
49.9 ± 11.8
0.062 ± 0.012
0.042 ± 0.008
46.5 ± 16.1
0.163 ± 0.017
0.106 ± 0.012
70.3 ± 33.3
0.139 ± 0.017
0.092 ± 0.011
61.9 ± 19.0
0.123 ± 0.016
0.082 ± 0.011
56.4 ± 11.9
0.135 ± 0.017
0.092 ± 0.012
32.0 ± 4.6
0.145 ± 0.016
0.094 ± 0.011
67.1 ± 19.9
0.160 ± 0.017
0.104 ± 0.012
37.6 ± 12.7
Ns, number of plants sampled per site; He, average expected heterozygosity (equivalent to Nei’s (1973) gene diversity); I, Shannon information index;
%p, percentage of polymorphic loci; PB, number of private bands; S, minimum number of leaves of flowering plants at a site; Mean(FPP), average number of flowers produced per plant at a site; NM, not measured; NS, not significant. *Sample size below 6, so the result has to be treated cautiously.
electropherograms for presence (1) or absence (0) of peaks. Analyses were confined to primer combinations that generated clearly readable profiles (Bonin et al., 2007). Manual scoring was repeated twice to estimate the effects of genotyping error. Double-reading of plants was repeated with an emphasis on individual identity in the case of inconsistency in individual banding patterns to ensure a high accuracy of band matching. The full AFLP analysis was repeated for 16 samples.
Statistical analysis of the genetic structure
The number of private bands (PB), average heterozygosity (He), Shannon information index (I) and percentage of polymorphic loci (%p) were computed for each population, assuming Hardy–Weinberg equilibrium using GenAlEx version 6.1 (Peakall & Smouse, 2006). Genetic diversity within and among populations was compared and regional differences in genetic variance were computed using the R 2.7.0 package (www.R-project.org).
To analyse population genetic structure, non-spatial Bayes- ian clustering was employed using structure version 2.2
(Pritchard et al., 2000; Falush et al., 2003). During structure simulations, individuals are probabilistically assigned to pop- ulations, or jointly to more populations if their genotypes indicate that they are admixtures (have mixed ancestry). An
‘admixture’ model was run with correlated allele frequencies,
50,000 Markov chain Monte Carlo (MCMC) iterations of burn-in period length to minimize the effects of the starting configuration and 50,000 MCMC replicates after burn-in for accurately assessing parameters. To estimate the K number of ancestral genetic populations and the ancestry membership proportions of each individual in these clusters, the algorithm was run 10 times for each user-defined K value from 2 to 20. The final K was estimated as the largest K value with a significantly higher likelihood than that from K ) 1 runs (two- sided nonparametric Wilcoxon test, P < 0.005; Pico´ et al.,
2008). The true number of K groups was also determined by plotting the mean likelihood of K and the rate of change of the likelihood distribution (mean ± SD) over 10 runs for each K value (Evanno et al., 2005). Membership fractions of individ- uals in the inferred clusters were computed and displayed using distruct (Rosenberg, 2004). arlequin version 2000
(Schneider et al., 2000) was used to carry out an analysis of molecular variance (AMOVA) to quantify population differ- entiation within and between regions, testing four models of population groups (Table 3). aflp-surv 1.0 (Vekemans et al.,
2002) was employed to produce a phylogenetic tree, computing FST values based on the Bayesian method with non-uniform priors (Zhivotovsky, 1999) as recommended by Bonin et al. (2007). neighbour and drawgram (both in the software package phylip version 3.67; Felsenstein, 2005) were then used to draw UPGMA (unweighted pair group method with arith- metic mean) dendrograms for the 1000 bootstrapped matrices. Isolation-by-distance patterns were investigated across all populations and at the group level to reveal any specific barriers to gene flow that might be present. Genetic distances among population pairs were computed using the Zhivotovsky (1999) method and the square-root method (Lynch & Milligan, 1994). Geographical distances between population pairs were calcu- lated as the shortest distance between two points on a sphere using the online calculator Cactus2000 (http://www. cactus2000.de/uk/unit/massgrk.shtml). ibd version 1.52 (Boho- nak, 2002) was employed to plot genetic distances as a function of geographical distances between population pairs, to test for significance in the isolation-by-distance relationship with a Mantel test for matrix correlation (1000 randomizations), and to determine the strength of this relationship by regressing all pairwise genetic similarity values against their corresponding
geographical distances using reduced major axis regression. Geographical trends in genetic and demographic structure
were analysed by computing the longitudinal gradient from distances between the longitude of a site and the average longitude of populations FR1 and FR3 (= 3.5), which repre- sent core sites, and the latitudinal gradient from distances between the latitude of a site and the average latitude of populations FR1 and FR3 (= 43.9). We tested for significant trends in genetic traits and demographic traits with increasing
Table 3 Analyses of molecular variance (AMOVA) testing four models of group formation. F statistics were computed from a matrix of Euclidean squared distances between every pair of individuals. While all populations were treated equally in model 1, models 2–4 tested different regional relationships (see Table 1 for assignment of populations to regions). All fixation indices (FSC, among populations and within groups; FST, among individuals within populations; FCT, among groups) were significant, except one in model 4.
Model 1, all populations; model 2, core, Northern, Southern, Atlantic, Eastern; model 3, South France versus South Spain; model 4, South France versus Atlantic group.
*1023 permutations, FCTnon-significant, P (random value > observed value) smaller than 0.10.
latitudinal and longitudinal distance from the distribution centre using linear modelling in the R 2.7.0 modelling package. The Wilcoxon test was used to check for significant differences in genetic and demographic traits between geographical regions using R 2.7.0.
Ten universal plastid markers (see Appendix S2) were screened for sequence variation among populations with proof-reading Taq using the ABI 377 DNA Automated Sequencer (Applied Biosystems). The primer combination accD 769F vs. psaI 75R (amplifying the accD–psa1 intergenic spacer) was able to produce sufficient variability between populations in form of multiple repeats (C)8to (C)11 followed by (T)8to (T)9between base pairs (bp) 618 and 637. A forward and a reverse primer (MSpsa1_F and MSpsa1_R) were designed for subsequent analysis of length variation at this locus for 211 plants: MSpsa1_F (5¢-AAG CAT CCC TCT CTT GAC AA-3¢) and MSpsa1_R (JOE labelled at 5¢- end) (5¢-CAA CAA ACA GGG ATT CCT AG-3¢).
Amplifications were performed in 10-lL reaction volumes with 0.4 lL extracted DNA, 0.2 lL of each of the primers,
0.2 lL bovine serum albumin (BSA) buffer and 0.2 lL of polymerase chain reaction (PCR) ready-mix (ABgene PCR Mastermix, 1.5 mm Mg, Abgene Limited, Epsom, UK). After a denaturation step at 94°C for 2 min, PCRs were performed with 30 cycles of 30 s denaturing at 93°C, 60 s of annealing at
48°C, 60 s of extension at 72°C, and a final 8-min extension step at 72°C. The PCR products were purified using QIAquick columns (Qiagen). Amplification success was confirmed on
1% agarose gels stained with ethidium bromide. The PCR products were diluted and run on an ABI Prism 310 Genetic Analyzer mixing 1 lL of the diluted sample with 10 lL of a formamide solution (High Dye) and 0.2 lL of GeneScanTM
500 ROXTMSize Standard (Applied Biosystems). Microsatellite
length variation at the MSpsa1 microsatellite locus was analysed to compute the relative proportion of haplotypes present in each population using GeneScan®Analysis Software version 2.0 and Genotyper® Software version 2.1 (Applied Biosystems Inc., Foster City, CA, USA). Distinct peaks were scored as present by visual inspection and the scoring procedure was repeated twice on separate occasions to minimize inconsistencies in scoring.
Geographical structure in demographic traits
NR and N differed among sites (Table 1). The number of flowers produced per plant varied strongly within and among sites (Table 2). There was a positive, significant influence of plant size on the number of flowers produced per plant at all sites except SA1 and SA2 (Table 2). The ratio NR/N decreased significantly with increasing latitudinal distance from the distribution centre (Fig. 2d). The mean number of flowers
Figure 2 Geographical trends of demographic traits of sampled populations of Himantoglossum hircinum in Europe. The range of the mean number of flowers produced per plant increased with increasing latitudinal (a) and longitudinal distance (b) from the distribution centre. The mean number of flowers produced per plant decreased significantly towards the eastern range margin when excluding two outlier populations, BW3 and BW5, from the linear model (c). The ratio NR/N decreased significantly with increasing latitudinal distance from the distribution centre (d).
produced per plant decreased significantly with increasing longitudinal distance towards the east when excluding two statistical outliers, BW3 and BW5, from the linear model
(P < 0.05, R2 = 0.421, Fig. 2c). The range of this parameter
increased considerably with increased longitudinal distance from the distribution core (Fig. 2a,b). The flowering percent- age and size of both populations, BW3 and BW5, were high despite their location at the eastern range margin.
Further significant differences among regions or geograph- ical trends (considering latitudinal or longitudinal distance from distribution centre, either in general, or towards the east/ west or the north/south) in demographic traits (considering population size, percentage of reproductive plants, mean number of flowers per plant) were not found.
The AFLP data matrix included 215 fragments (mean fragment size: 144 ± 62). Of these, 205 (95.35%) were polymorphic. Fragment size and frequency were not correlated. Genotyping error was negligible for microsatellite analyses, because peaks were clearly readable for all 211 samples and double-reading them resulted in 100% accuracy. Selective amplification repeated for 16 AFLP samples produced identical banding patterns. For 11 samples, all steps were repeated (starting from DNA extraction), and genotyping error was 6% (using only 120 loci that were absolutely clear) and 11% (using all
Geographical genetic structure
Five haplotypes (H1 to H5) were recognized across all populations based on the MSpsa1 locus. Haplotype number (Nhaplo) differed strongly among sites. The two core popula- tions, FR1 (Nhaplo= 4) and FR3 (Nhaplo= 5), had significantly more haplotypes compared with all other populations (Wilcoxon test, P < 0.05). Haplotype frequencies differed between geographical areas (Fig. 1). H1 dominated in Atlantic (78.6%) and northern (80.0%) populations, H2 (31.8%) and H3 (54.6%) in southern Spain (50%) and H4 in Italy.
Populations differed strongly in their genetic diversity, with high intraregional variability at the range margins (Table 2). A few populations had no exclusive bands and many had just one or two exclusive bands. The Italian population had 11 private
bands. Average genetic diversity was lower in northern (not
significant) and eastern peripheral populations (significant, Wilcoxon test, P < 0.05) compared with that in core popu- lations, but was only slightly lower for southern peripheral populations. Atlantic populations exhibited relatively high genetic diversity. When excluding outlying populations from the analyses, geographical trends in genetic diversity became significant (Fig. 3a,b).
All fixation indices were significant except for differentiation among groups in model 4 (Table 3). Molecular variance within populations was much higher than variance among popula- tions, resulting in a fixation index of FST= 0.146 (Model 1). At the regional level, genetic differentiation increased from the geographical range centre to the geographical range margins (Table 2). Genetic differentiation among Atlantic populations
Figure 3 Geographical trends of genetic diversity (He) of Himantoglossum hircinum populations within its European distribution. (a) Mean genetic diversity plotted against increasing latitudinal distance from the distribution centre. (b) Mean genetic diversity plotted against increasing longitudinal distance from the distribution centre. Dashed lines represent predictions of linear models when including all populations; black lines when excluding statistical outliers.
was low even when grouping them together with populations SE1 and SE4 at the northern range margin (FCT = 0.028, not significant). Southern French populations were not strongly differentiated from southern Spanish populations or Atlantic populations (Table 3). Analysed over all populations, genetic distance increased significantly with geographical distance (Lynch & Milligan: r = 0.27, P < 0.05; Zhivotovsky: r = 0.31, P < 0.01). No correlations were detected between genetic diversity (e.g. diversity indices, haplotype frequency) and population demographic traits.
The phylogenetic tree did not reveal major groups in Europe. However, most populations at the eastern range margin clustered together except for SA2 and BW2 (Fig. 4). The Italian site was strongly differentiated from all other sites. Six ancestral genetic clusters (C1–C6, Wilcoxon test P < 0.05) were inferred with structure (‘admixture’, see Appendix S3). Membership fractions of individuals in the six clusters varied between seven geographical regions (Fig. 5). Cluster three (C3) had high membership proportions in southern Spain and the Atlantic enclave (Fig. 5). Central Germany showed member- ship fractions that differed markedly from other sites within the eastern range margin (see Appendix S4). Cluster five (C5) was most pronounced in southern France and southern Spain. French and English sites were characterized by high member- ship fractions in cluster one (C1). The eastern range margin and the Swiss population shared substantial membership fractions in all clusters. Cluster six (C6) presented high proportions at the eastern and northern margins and at the Swiss site (> 59%), intermediate proportions in southern France and the Atlantic group (38% and 39%), but low to very low proportions in southern Spain and Italy (Fig. 5). The Italian site was the only site where cluster two (C2) and cluster four (C4) had high membership fractions.
DI SCUSSI ON
Geographical population structure
Decreased habitat suitability towards the range margins is expected to reduce population size and increase isolation among populations, resulting in the loss of genetic diversity and net flux of genes from central to marginal sites (‘swamp- ing’; Kirkpatrick & Barton, 1997). While the reduced genetic diversity and haplotype frequency at many eastern marginal sites of H. hircinum matches expectations for the genetic consequences of historical range expansions (Hewitt, 1996), the decrease in genetic diversity towards the range margins of the species is not consistent. Some sites at the southern periphery (IT3, SP1) and within the Atlantic group were among the populations with highest genetic diversity (He > 0.20). We could detect significant trends in genetic diversity along geographical gradients after excluding outlying populations (Fig. 3a,b). Population history and different processes limiting population growth at different boundaries of a species (Bullock et al., 2000) may be more important for the genetic structure of those outliers. This might also explain